fix: Comments to English for better dev experience (#15768)
Signed-off-by: yihong0618 <zouzou0208@gmail.com>
This commit is contained in:
@@ -509,7 +509,7 @@ class DeepseekVLV2ForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
|
||||
_, hw, n_dim = images_embeds.shape
|
||||
h = w = int(hw**0.5)
|
||||
|
||||
# 根据self.tile_tag & self.global_view_pos填充image token sequence
|
||||
# fill image token based on self.tile_tag & self.global_view_pos
|
||||
tile_index = 0
|
||||
vision_embeddings = []
|
||||
for jdx in range(images_spatial_crop.size(0)):
|
||||
|
||||
Reference in New Issue
Block a user